karps alternatives and similar packages
Based on the "Big Data" category.
Alternatively, view karps alternatives based on common mentions on social networks and blogs.
Do you think we are missing an alternative of karps or a related project?
Karps-Haskell - Haskell bindings for Spark Datasets and Dataframes
This project is an exploration vehicle for developing safe, robust and reliable data pipelines over Apache Spark. It consists in multiple sub-projects:
- a specification to describe data pipelines in a language-agnostic manner, and a communication protocol to submit these pipelines to Spark. The specification is currently specified in this repository, using Protocol Buffers 3 ( which is also compatible with JSON).
- a serving library, called karps-server, that implements this specification on top of Spark. It is written in Scala and is loaded as a standard Spark package.
- a client written in Haskell that sends pipelines to Spark for execution. In addition, this client serves as an experimental platform for whole-program optimization and verification, as well as compiler-enforced type checking.
There is also a separate set of utilities to visualize such pipelines using Jupyter notebooks and IHaskell.
This is a preview, the API may (will) change in the future.
The name is a play on a tasty fish of the family Cyprinidae, and an anagram of Spark. The programming model is strongly influenced by the TensorFlow project and follows a similar design.
Karps can also take advantage of the Haskell kernel for Jupyter, which provides a better user experience and comes with beautiful introspection tools courtesy of the TensorBoard server. Using Tensorboard, you can visualize, drill down, introspect the graph of computations:
Some notebooks that showcase the current capabilities are in the
directory. Some prerendered versions are also available. Chrome seems to provide
the best experience when playing interactively with the visualizations.
Installation (for users)
These instructions assume that the following software is installed on your computer:
- Spark 2.x (2.1+ strongly recommended). See the installation instructions for a local install. It is usually a matter a downloading and unzipping the prebuilt binaries.
- the stack build tool
Launching Spark locally Assuming the
SPARK_HOME environment variable is set
to the location of your current installation of Spark, run:
$SPARK_HOME/bin/spark-shell --packages krapsh:karps-server:0.2.0-s_2.11\ --name karps-server --class org.karps.Boot --master "local" -v
You should see a flurry of log messages that ends with something like:
WARN SparkContext: Use an existing SparkContext, some configuration may not take effect. The server is now running.
Connecting the Karps-Haskell client All the integration tests should be able to connect to the server and execute some Spark commands:
stack build stack test
You are now all set to run your first interactive program:
import Spark.Core.Dataset import Spark.Core.Context import Spark.Core.Functions let ds = dataset ([1 ,2, 3, 4]::[Int]) let c = count ds createSparkSessionDef defaultConf mycount <- exec1Def c
Installation (GUI, for users)
IHaskell can be challenging to install, so a docker installation script is provided. You will need to install Docker on your computer to run Karps with IHaskell.
In the project directory, run:
docker build -t ihaskell-karps . docker run -it --volume $(pwd)/notebooks:/karps/notebooks \ --publish 8888:8888 ihaskell-karps
notebooks directory contains some example notebooks that you can run.
Note that it still requires a running Spark server somewhere else: the docker container only runs the Haskell part.
MacOS users When running Docker with OSX, you may need to tell Docker how to communicate from inside a container to the local machine (if you run Spark outside Docker). Here is a command to launch Docker with the appropriate options:
docker run -it --volume $(pwd)/notebooks:/karps/notebooks \ --publish 8888:8888 --add-host="localhost:10.0.2.2" ihaskell-karps
Standalone linux installation The author cannot support the vagaries of operating systems, especially when involving IHaskell, but here is a setup that has shown some success:
In Ubuntu 16.04, install all the requirements of IHaskell (libgmp3-dev ghc ipython cabal-install, etc.)
kraps-haskell directory, run the following commands:
export STACK_YAML=$PWD/stack-ihaskell.yaml stack setup 7.10.2 # This step may be required, depending on your version of stack. # You will see it if you encounter some binary link issues. stack exec -- ghc-pkg unregister cryptonite --force stack update stack install ipython-kernel-0.8.3.0 stack install ihaskell-0.8.3.0 stack install ihaskell-blaze-0.3.0.0 stack install ihaskell-basic-0.3.0.0 stack install ihaskell install --stack stack exec --allow-different-user -- jupyter notebook --NotebookApp.port=8888 '--NotebookApp.ip=*' --NotebookApp.notebook_dir=$PWD
This project has so far focused on solving the most challenging issues, at the expense of breadth and functionality. That being said, the basic building blocks of Spark are here:
- dataframes, datasets and observables (the results of
- basic data types: ints, strings, arrays, structures (both nullable and strict)
- basic arithmetic operators on columns of data
- converting between the typed and untyped operations
- grouping, joining
You can take a look at the notebooks in the
notebooks directory to see what is
What is missing? A lot of things. In particular, users will most probably miss:
- an input interface. The only way to use the bindings is currently to pass a list of data.
- long types, floats, doubles
- broadcasting observables (scalar * col). This one is interesting and is probably the next piece.
- setting the number of partitions of the data
Contributions are most welcome. This is the author's first Haskell project, so all suggestions regarding style, idiomatic code, etc. will be gladly accepted. Also, if someone wants to setup a style checker, it will be really helpful.
The API and design goals are slightly more general than Spark's. A more thorough
explanation can be found in the